-
Notifications
You must be signed in to change notification settings - Fork 0
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Fix usage of as_offset #542
Conversation
cscs-ci run default |
launch jenkins spack |
Mandatory Tests Please make sure you run these tests via comment before you merge!
Optional Tests To run benchmarks you can use:
To run tests and benchmarks with the DaCe backend you can use:
In case your change might affect downstream icon-exclaim, please consider running
For more detailed information please look at CI in the EXCLAIM universe. |
cscs-ci run default |
1 similar comment
cscs-ci run default |
launch jenkins spack |
Fix usage of as_offset in following stencils: * _compute_hydrostatic_correction_term * _compute_horizontal_gradient_of_exner_pressure_for_multiple_levels * _truly_horizontal_diffusion_nabla_of_theta_over_steep_points --------- Co-authored-by: Edoardo Paone <[email protected]> Co-authored-by: Magdalena Luz <[email protected]>
The lowering of
as_offset
in GT4Py was wrong, which was discovered and fixed recently in GridTools/gt4py#1484. Apparently we have some field operators in icon4py that just massaged the code in such a way that the wrong behavior in GT4Py was "canceled" out. However with the fix in GT4Py this broke and and the type inference on ITIR rightfully complained with an error similar toThis PR is meant as a starting point to fix the remaining stencils (e.g.
_truly_horizontal_diffusion_nabla_of_theta_over_steep_points
looks like a candidate). Note that we can currently not detect these issues in the frontend as the experimentalas_offset
has a conceptual problem preventing this which can be understood from this snippet:I very much suspect field operators that use values of the erroneous field operators, e.g. combined (formerly called fused) field operators, and shift those values have produced incorrect results in the past. For the person taking over this PR please be very careful and validate that my changes below.